Hi,

that could be the case, the data is not small but has probably many
duplicates.

thanks!

On Fri, Apr 6, 2012 at 7:21 PM, Kenneth Heafield <[email protected]>wrote:

> Hi Eva,
>
>        The smoothing is out of whack, probably because your training data
> is
> too small or contains an unnatural number of duplicates.
>
> Kenneth
>
> On 04/06/2012 02:14 PM, Eva Hasler wrote:
> > Hi,
> >
> > I'm training an srilm language model (English) with the following
> > settings srilm/bin/i686-m64/ngram-count -order 5 -interpolate
> > -kndiscount -unk
> >
> > and the error is this:
> > "one of modified KneserNey discounts is negative
> > error in discount estimator for order 4"
> >
> > I've not seen this error before, anybody knows what it could be?
> >
> > Eva
> >
> >
> >
> > _______________________________________________
> > Moses-support mailing list
> > [email protected]
> > http://mailman.mit.edu/mailman/listinfo/moses-support
> _______________________________________________
> Moses-support mailing list
> [email protected]
> http://mailman.mit.edu/mailman/listinfo/moses-support
>



-- 
@__^.^
_______________________________________________
Moses-support mailing list
[email protected]
http://mailman.mit.edu/mailman/listinfo/moses-support

Reply via email to